The dual role of weather forecasts on changes in
activity-travel behavior
Mario Coolsa,b,c, Lieve Creemersb
aUniversit de Li`ege, TLU+C (Transport, Logistique, Urbanisme, Conception), Chemin
des Chevreuils 1 B.52/3, 4000 Li`ege, Belgium
bTransportation Research Institute (IMOB), Hasselt University, Wetenschapspark 5, bus
6, 3590 Diepenbeek, Belgium
cHogeschool-Universiteit Brussel, Centre for Information, Modelling and Simulation,
Warmoesberg 26, 1000 Brussels, Belgium
Abstract
A deeper understanding of how human activity-travel behavior is affected by various weather conditions is essential for both policy makers and traffic managers. To unravel the ambiguity in findings reported in the literature, the main objective of this paper is to obtain an accurate assessment of how weather forecasts trigger changes in Flemish activity-travel behavior. To this end, data were collected by means of a stated adaptation experiment, which was administered both on the Internet and via traditional paper-and-pencil questionnaires. To address the main research question of this paper, two statistical techniques were adopted. The first technique is the computation of Pearson chi-square independence tests. The second approach is the esti-mation of a GEE-MNL-model. The results from both techniques underscore the dual role of weather forecasts on changes in activity-travel behavior. On the one hand, the results clearly illustrate the significant effect of forecasted weather; the likelihood of changes in activity-travel behavior significantly de-pends on the weather forecasted. On the other hand, different methods of acquiring weather information (exposure, media source, or perceived relia-bility) do not impact the probability of behavioral adaptations. This duality may be partially attributable to the discrepancy that exists between weather forecasts and true traffic and roadway conditions. Therefore, the implemen-tation of a road weather information system that is directly linked to the weather forecasts is recommended.
1. Introduction
1
As discussed by Cools et al. (2010a), a deeper understanding of how
2
various weather conditions affect human activity-travel behavior is essential
3
for policy makers and traffic managers. It provides insights that might help
4
alleviate negative effects of the road network that are often associated with
5
adverse weather. A multitude of changes in activity-travel behavior can be
6
triggered by different weather conditions. These include (i) trip cancelations
7
(elimination of the activity from the activity agenda) (e.g., Madre et al.
8
(2007), Wilton et al. (2011), Kim et al. (2010)); (ii) changes in the location
9
where the activity is performed (e.g., Hagens (2005), Koetse and Rietveld
10
(2009)); (iii) changes in the timing of the activity or the corresponding trip
11
(e.g., Chung et al. (2005), Maze et al. (2006)); (iv ) changes in the transport
12
mode (e.g., Akar and Clifton (2009), Guo et al. (2007), Kuhnimhof et al.
13
(2010)); and (v ) changes in the route for the trip (e.g., Lam et al. (2008),
14
Sumalee et al. (2011)).
15
In addition to actual weather conditions, weather reports and forecasts
16
(information on current and future weather conditions) influence travel
be-17
havior. In this regard, it is worth consulting reports regarding traffic
informa-18
tion provision, for instance, advanced traveler information systems (ATIS).
19
These systems have the potential to increase the efficiency of transportation
20
systems as well as their usefulness to individual travelers (Wang et al., 2009).
21
The provision of traffic information can induce a similar range of changes
22
in activity-travel behavior as the responses to different weather conditions
23
(e.g., Rodr´ıguez et al. (2011), Casas and Kwan (2007), Son et al. (2011), Son
24
et al. (2011), Tseng et al. (2012)). Nonetheless, the impact of the provided
25
information should not be overestimated, as the perceived value of
acquir-26
ing information is often limited (Chorus et al., 2006b; Lyons, 2006). The
27
success of information provision is contingent on the characteristics of the
28
information itself, such as its quality, accuracy, usefulness, timeliness, cost,
29
and communication mode (Zhang and Levinson, 2008). Moreover,
socio-30
economic and contextual variables significantly influence the impact of the
31
information provision (Joh et al., 2011; Chorus et al., 2006a; Ben-Elia et al.,
32
2008).
33
The published literature regarding the impact of information (weather
34
forecasts) on activity-travel behavior is ambiguous. Khattak and de Palma
35
(1997) reported that forecasted weather information did not significantly
36
affect the probabilities of adapting mode and departure time. In contrast,
the studies by Hagens (2005), Sihvola (2009), and Kilpel¨ainen and Summala
38
(2007) demonstrated significant impacts. Thus, a fundamental question is
39
whether forecasted weather information triggers changes in activity-travel
40
behavior. This paper focuses on the impact of weather forecasts on
activity-41
travel behavior.
42
An important issue in the cross-national transferability of findings is the
43
fact that activity-travel behavior varies across spatial and temporal contexts
44
(Khattak and de Palma, 1997). Consequently, published results and
dis-45
cussions are not always applicable to specific regional context(s), such as
46
Flanders, the northern part of Belgium, which is the regional context
con-47
sidered in the present paper. Take, for example, the results of de Palma and
48
Rochat (1999) and Kilpel¨ainen and Summala (2007), which were obtained
49
in Switzerland and Finland, respectively. Adverse weather conditions such
50
as snow and hail occur more frequently in these countries than in Flanders.
51
It can be assumed that because of habituation, people in these countries
52
experience these weather phenomena differently and, therefore, adapt their
53
activity-travel behaviors differently.
54
In addition, most weather-related studies make no differentiation based
55
on the particular activity. This is a shortcoming in the literature because
56
people are less likely to change their regular commuting behavior due to
57
weather forecasts than they are to alter trips for non-work/school-related
58
purposes. Moreover, the majority of these studies only focus on a subset of
59
weather types, mostly rain and snow.
60
Given the above considerations, the main objective of this paper is to
ac-61
curately assess how weather forecasts change Flemish activity-travel
behav-62
iors, taking into account the full context of behavioral adaptations, activity
63
purposes and weather types to clarify the ambiguities in published results and
64
to verify the transferability of the results of previous studies to the context
65
of Flanders.
66
2. Data
67
2.1. Stated adaptation experiment
68
Data regarding the impact of weather forecasts on Flemish activity-travel
69
behaviors were collected by means of a stated adaptation experiment, which
70
was carried out in March and April of 2009. Respondents, who were recruited
71
by means of convenience sampling, were asked to indicate if and how they
would change their activity-travel behaviors considering various experimental
73
attribute profiles corresponding to different weather conditions.
74
In total, 586 respondents completed the stated adaptation survey, which
75
was administered both on the Internet (86.7%) and via traditional
paper-76
and-pencil questionnaires (13.3%). As documented by Cools et al. (2010a),
77
this dual-mode administration was chosen to remedy the sample bias that
78
can be introduced when only internet-based data collection is conducted. In
79
total, 90 behavioral adaptations in response to different weather conditions
80
were queried; the frequencies of 5 travel behavior changes in response to 6
81
weather conditions were determined, and this was repeated for 3 types of
82
trips. These 90 behavioral adaptations were assayed for both actual weather
83
and forecast weather conditions, resulting in a final total of 180 potential
84
behavioral adjustments.
85
The three types of trips considered correspond to the categories of most
86
commonly performed trips according to the Flemish travel behavior survey
87
(Cools et al., 2010b): commuting (work/school), shopping and leisure trips.
88
For each of these types of trips, the respondents indicated how often (never,
89
in 1-25% of the cases, in 26-50% of the cases, or in more than 50% of the
90
cases) they would make a certain change in activity-travel behavior. The
91
following changes in travel behavior were queried: (i) changing the transport
92
mode, (ii) changing the timing of the trip (postponing or advancing the trip
93
to a later/earlier time on the same day), (iii) changing the location of the
94
activity (work/school, shopping or leisure), (iv) canceling the trip altogether,
95
and (v) changing the route of the trip.
96
In accordance with Cools et al. (2010c), who identified the weather
con-97
ditions that had significant impacts on the daily traffic intensities of Belgian
98
highways, the following weather conditions were considered: cold
tempera-99
tures (defined as temperatures below freezing (0◦C, 32◦F), abbreviated as 100
‘cold’), warm temperatures (defined as temperatures above 28◦C (82.4◦F), 101
abbreviated as ‘warm’), snow/freezing rain, heavy rain/thunderstorms
(ab-102
breviated as ‘rain’), fog and storms/heavy wind. The question format is
103
illustrated in Figure 1.
104
In addition to the stated adaptation questions, the survey also
explic-105
itly queried information concerning the average exposure of respondents to
106
weather forecasts in their daily lives. In particular, the frequency of this
107
exposure was ascertained as well as the media source(s) involved and the
108
perceived reliabilities of the weather conditions forecast (measured on a
10-109
point scale). Furthermore, the survey collected information concerning the
Do you postpone or advance your work/school-related trip to a later/earlier moment the same day due to any of the following forecasted weather conditions?
Mark the answer that corresponds mostly to your situation. Only one answer is possible for each forecasted weather condition.
No, never Yes, occasionally Yes, sometimes Yes, usually (<25% of (<50% of (>50% of the cases) the cases) the cases)
Cold temperature Snow/freezing rain Heavy rain/thunderstorm Fog Warm temperature Storm/heavy wind
Figure 1: Stated adaptation question concerning postponing/advancing work/school-related trips
respondents’ socio-demographic profiles and queried different activities and
111
trip-related attributes. Although a convenience sample was used for this
112
study, the respondents’ age, gender and marital state were used as the
ba-113
sis for calculating weights that guarantee optimal correspondence between
114
the survey sample composition and the Flemish population. The weights
115
were calculated by matching the relative frequencies of the three-way
cross-116
tabulations of the sample with those of the total population. Because all the
117
cross-tabulations were known, such that the multivariate correlations were
118
taking into account, the weights ensured that the relative frequencies of the
119
weighted sample corresponded exactly to those of the total population. It is
120
worth noting that all the tables and figures presented in this paper are based
121
on the weighted results.
2.2. Data description
123
Recall that the main goal of this paper is to investigate how weather
124
forecasts trigger changes in Flemish activity-travel behavior. The study was
125
based on the following five specific research questions:
126
1. Do changes in activity-travel behavior depend on forecasted weather
127
conditions?
128
2. Do changes in activity-travel behavior depend on degrees of exposure
129
to weather forecasts?
130
3. Do changes in activity-travel behavior depend on the media sources of
131
weather forecasts?
132
4. Do changes in activity-travel behavior depend on the perceived
relia-133
bility of weather forecasts?
134
5. Which factors trigger changes in activity-travel behavior in the
pres-135
ence of adverse weather conditions, and, in particular, what roles are
136
played by the weather-forecast characteristics (exposure, media source,
137
perceived reliability) considered herein?
138
The dependent variables required to tackle the first four questions are
139
the changes in activity-travel behavior in response to the forecasted weather
140
conditions. As mentioned in the previous subsection, 90 behavioral changes
141
were queried with regard to weather forecasts. These changes in
activity-142
travel behavior are displayed in Figure 2. Note that the original response
143
categories (never, in 1-25% of the cases, in 26-50% of the cases, or in more
144
than 50% of the cases) have been dichotomized to increase the interpretability
145
of the graph as well as for the methodological reasons discussed in Section 3.
146
From Figure 2, one can clearly see that travelers do adapt their activity-travel
147
patterns in response to forecasted weather conditions. This is especially
148
the case for trips with non-mandatory activity-trip purposes (i.e., shopping
149
and leisure trips). The forecasted weather condition that appears to trigger
150
the most changes is snow, while cold weather had the least impact. The
151
remarkably strong behavioral reactions to forecasted storms are in line with
152
the published literature regarding actual weather effects (e.g., Cools et al.
153
(2010a) and Cools et al. (2010c)).
154
To address the fifth research question, we have investigated behavioral
155
changes in response to ‘actual’ weather forecasts, taking into account the
156
different features of weather forecasts. Because respondents could indicate
157
multiple changes in activity-travel behavior simultaneously, it was necessary
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Cold Snow Rain Fog Warm Storm Mode Change Postpone Trip Location Change Cancelation Trip Route Change 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Cold Snow Rain Fog Warm Storm Mode Change Postpone Trip Location Change Cancelation Trip Route Change 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Cold Snow Rain Fog Warm Storm Mode Change Postpone Trip Location Change Cancelation Trip Route Change
Figure 2: Behavioral changes in response to different forecasted weather conditions
to prioritize the different changes. The main selection criterion behind this
159
prioritization is the overall impact of a given change on the activity-travel
be-160
havior from an environmental perspective. Note that a comparable approach
161
was followed by Cools et al. (2011) in their assessment of the impact of road
162
pricing on changes in activity-travel behavior. The prioritization scheme is
163
displayed in Table 1.
The following example clarifies this scheme. A respondent indicated that,
165
in response to heavy rain, he/she never changes the travel route, changes the
166
activity location or makes trip changes in 1-25% of the cases, and alters
167
his/her transport mode or the timing of the trip 26-50% of the time. For
168
this respondent, the ranks for a mode change, time-of-day change, location
169
change, trip cancelation and route change are 8, 13, 7, 6 and 16, respectively.
170
The action corresponding to the lowest rank – in this case, trip cancelation
171
(which has a rank value of 6) – is the adaptation considered for the modeling
172
process because this adaptation is likely to have the largest impact from
173
an environmental point of view. If the respondent did not consider any
174
change(s), then all changes have a rank value of 6 and, correspondingly, ‘No
175
change’ would be the respondent’s choice option. Table 2 displays the overall
176
percentages of this prioritized adaptation variable. In agreement with results
177
related to weather forecasts (Figure 2), more behavioral changes are made
178
when discretionary trips are involved.
179
Table 1: Prioritization of the behavioral adaptation
Mode Time of Day Location Trip Route Change Change Change Cancelation Change
Never 16
0-25% 9 15 7 6 14
26-50% 8 13 4 3 12
>50% 5 11 2 1 10
In addition to dependent variables, different explanatory variables are
180
used to investigate the impact of weather forecasts. For the continuous
pre-181
dictors, mean and standard deviation values are provided, while, for
categor-182
ical variables, the percentages of each class are tabulated and the reference
183
category is highlighted. These categorical variables are internally coded as
184
(k − 1) dummy (0-1) variables, where k is the number of classes.
185
The first group of explanatory variables corresponds to the key
vari-186
ables in this study, which are the following weather-forecast-related variables:
187
forecasted weather conditions, average exposure to weather forecasts, media
188
source and perceived reliability of the weather forecast. As shown in Table
189
2, approximately 60% of the respondents absorb weather information on a
190
daily basis. The most important media sources for weather information are
191
television and, to a lesser extent, radio. The internet and newspapers clearly
192
play smaller roles. In general, the respondents appear to be satisfied with
193
the reliability of forecasted weather information.
Table 2: Descriptive statistics Variable name Description
Dependent variables: Prioritized behavioral adaptation
Work/school Mode: 10.2%, Time-of-day: 10.0%, Location: 3.4%, Cancelation: 8.5%, Route: 8.2%, No change1: 59.7%
Shopping Mode: 5.5%, Time-of-day: 5.5%, Location: 9.5%, Cancelation: 33.1%, Route: 1.2%, No change1: 45.2%
Leisure Mode: 5.4%, Time-of-day: 4.2%, Location: 8.2%, Cancelation: 33.8%, Route: 1.8%, No change1: 46.6%
Independent variables
Weather forecast characteristics
Weather type2 Cold: 16.7%, Snow: 16.7%, Rain: 16.7%,
Fog: 16.7%, Warm1:16.7%, Storm:16.7%
Exposure Daily1: 59.3%, Weekly: 33.2%, Occasionally: 7.5%
Media source3 Television1: 81.2%, Radio: 63.4%, Internet: 23.1%, Newspaper: 22.9%
Perceived reliability Low (1-5): 15.4%, High1(6-10): 84.6%
Socio-demographic characteristics
Age Mean: 43.3, Standard Deviation: 15.1 Gender Female: 49.0%, Male1: 51.0%
Children No1: 35.3%, Yes: 64.7%
Degree No secondary: 12.9%, Secondary: 36.3%, College: 30.5%, University1: 20.2%
Income Low1 (≤ 1250e): 20.2%, Medium-High (> 1250e): 70.0%, Unspecified: 9.7%
Marital state Single1: 36.7%, Married: 63.3%
Profession Professionally active: 73.3%, Students1: 11.3%, Inactive: 15.4%
Urbanization Metropolitan: 16.5%, Strong: 11.3%, Moderate: 50.8%, Weak1: 21.5%
Transport-related characteristics
Bicycle ownership No: 6.5%, Yes1: 93.5%
Car ownership No: 3.1%, Yes1: 96.9%
Driving license No: 11.6%, Yes1: 88.4%
Season ticket No: 60.9%, Yes1: 39.1%
Work/school trips Mean: 4.4, Standard Deviation: 3.7 Shopping trips Mean: 2.1, Standard Deviation: 1.7 Leisure trips Mean: 3.7, Standard Deviation: 2.7
1: Reference category
2: The percentages are equal because of the experimental design
3: The percentages do not sum to 100% because the 4 media sources were queried separately
In addition to weather-forecast-related variables, the explanatory
vari-195
ables include different descriptors of the socio-demographic profiles of
re-196
spondents. Recall that the results in this paper are based on weighted
re-197
sults, which is also the case for the descriptive statistics displayed in Table
198
2. The following variables were considered: age, gender; children, degree,
199
income, marital state, profession and urbanization. In addition,
related variables were considered; in particular, bicycle and car ownership
201
within the household as well as possession of a driving license and/or a
pub-202
lic transport season ticket were envisaged. In addition to variables related
203
to the availabilities of transport options, actual travel behavior was surveyed
204
by recording the weekly frequency of work/school trips, shopping trips and
205
leisure trips.
206
3. Methodology
207
To address the main research question of this paper, two statistical
tech-208
niques were adopted. The first technique, the Pearson chi-square
indepen-209
dence test, was used to test the first four research questions. This technique
210
was adopted to assess the (univariate) relationship between weather forecast
211
attributes (i.e., exposure, media source and perceived reliability) and changes
212
in response(s) to forecasted weather conditions.
213
The Pearson statistic Qp is defined by Equation 1:
Qp = k X i=1 l X j=1 (nij − ˆµij)2 ˆ µij , (1)
where nij is the observed frequency in cell (i, j), which is calculated by mul-214
tiplying the observed chance by the sample size, and ˆµij is the expected 215
frequency for table cell (i, j). When the row and column variables are
inde-216
pendent, Qp has an asymptotic chi-square distribution with (k − 1)(l − 1) 217
degrees of freedom (Agresti, 2002). The test assumes that at least 80% of
218
the cells have expected frequencies of 5 or more. When this assumption is
219
not met, modifications of the answer categories are required to ensure that
220
this criterion is satisfied. This is operationalized in the present study by
221
reducing the original response categories (never, in 1-25% of the cases, in
222
26-50% of the cases, or in more than 50% of the cases) to the dichotomous
223
answer possibilities ‘Change’ and ‘No change’.
224
Secondly, to investigate the fifth research question (the identification of
225
factors that trigger changes in activity-travel behavior in the presence of
ad-226
verse weather conditions and the determination of the role of weather-forecast
227
characteristics), a GEE-MNL-model was constructed. The GEE-MNL model
228
extends the classical multinomial logit (MNL) model by explicitly taking into
229
account the correlated responses by means of a marginal effect model that is
estimated using generalized estimating equations (GEE). In marginal
mod-231
els, the mean function is modeled directly, and the correlation structure is
232
regarded as a nuisance parameter. It is important to consider this
corre-233
lation structure, as behavioral adaptations in response to different weather
234
conditions are most likely correlated. In other words, a certain behavioral
235
adaptation in response to one weather condition is likely to be correlated to
236
the behavioral adaptation in response to another weather condition.
237
To estimate the GEE-MNL model, the procedure suggested by Kuss and
238
McLerran (2007) was followed: the GEE-MNL model was specified as a
239
marginal model by reorganizing the response vector in a way that enables it
240
to be fitted as a multivariate binary model. The original variable Yij, corre-241
sponding to behavioral adaptations in response to certain weather variables,
242
is now written as an ((R − 1) × 1)-vector Y?
ij of binary variables Yijr? , such 243
that Yij = 2, ..., R results in Yij? = 1 in column r and 0 anywhere else. 244
In the case of Yij = 1 (reference category), Yij? = 0 in all R − 1 columns. 245
In the present paper, R equals 6 (5 behavioral changes + the no-change
al-246
ternative in cases where no change was made), and the no-change alternative
247
is defined as the reference category.
248 Let Y? i = (Y? 0 i1, ..., Y? 0
in1) denote the (ni(R − 1) × 1) response vector for the
i-th cluster with expectation π?
i and covariance matrix Vi?. This covariance
V?
i is a ‘double-block’ diagonal matrix, where the (R − 1) × (R − 1)-block
for (r,r0) on the ‘inner’ block of the main diagonal of V?
i is a multinomial
covariance matrix for the j-th observation in the i-th cluster. Furthermore, it is noteworthy that the remaining elements on the ‘outer’ block specify the covariance between two different observations (j,j0) in the i-th cluster.
Formally, this amounts to
V∗ i = cov(Yijr∗ , Y∗ij0r0) = π∗ ijr(1 − π∗ijr) if j = j0, r = r0 −π∗ ijrπijr∗ 0 if j = j0, r 6= r0 corr(Y∗ ijr,Y∗ij0r0) √ π∗
ijr(1−π∗ijr)π∗ij0r0(1−π ∗ ij0r0)
if j 6= j0
, (2)
where the first two lines of Equation 2 correspond to the ‘inner’ block of V? i ,
the third line corresponds to the ‘outer’ block, and π∗
ijr = E[Yijr∗ = 1]. It
should be noted that the third line does not constitute a circular definition. Instead, corr(Y∗
ijr, Yij∗0r0) must be given a working correlation pattern in the
equation: log( π ? ir 1 − π? ir ) = θ? r+ Xij0 βr?, (3) where π?
ir denotes the expectation of all elements of Yi? belonging to response 249
category r, θ?
r a vector of parameters to be estimated and Xij the vector of 250
explanatory variables. Note that there is no reference to a random effect in
251
the model equation.
252
4. Results
253
4.1. Impact of forecasted weather conditions and weather forecast
character-254
istics
255
Remember that, to address the first four research questions, the different
256
behavioral changes (displayed in Figure 2) were analyzed using independence
257
tests. Thus, for the first four research questions the 5 behavioral changes were
258
all taken into account and not prioritized as is the case for the fifth research
259
question. Table 3 displays the results of the statistical tests assessing the
260
dependence of behavioral changes on the type of forecasted weather. This
261
table indicates that all behavioral changes depend (with statistical
signifi-262
cance) on the type of forecasted weather. In other words, across the three
263
different types of trips and across the six behavioral changes, the type of
264
forecasted weather clearly influences the changes in travelers’ activity
pat-265
terns. Take, for example, the dependence of mode changes in work/school
266
trips on the type of forecasted weather. The chi2-value for this test equals
267
108.95, and the degrees of freedom equal 15 (=(6 weather conditions - 1)×
268
(4 response levels - 1), which yields a p-value that is smaller than 0.001 and,
269
thus, smaller than the typical level of significance (α = 0.05). Consequently,
270
the null hypothesis of this test, that mode changes in work/school trips are
271
independent of the type of forecasted weather, should be rejected.
Further-272
more, it can be concluded that the type of forecasted weather significantly
273
influences mode changes in work/school trips. Similar conclusions can be
274
drawn for all the other tests presented in Table 3.
275
With regard to the effect of the travelers’ average exposure to weather
276
forecasts, one can discern from Table 4 that the exposure level to weather
277
forecasts has only limited influence on the changes in activity-travel behavior.
278
With regard to work/school trips, the chi2-tests indicate that, regardless
279
of the weather condition considered, the behavioral changes in response to
280
weather forecasts do not significantly depend on the exposure level(s) to
Table 3: Dependence of behavioral changes on the type of forecasted weather Behavioral Change Work/School Shopping Leisure
Chi2 DF Chi2 DF Chi2 DF
Mode Change 108.95 15 57.58 15 80.03 15
Postpone Trip 353.43 15 290.50 15 311.48 15 Location Change 51.51 5 101.01 15 42.55 15 Cancelation Trip 111.28 5 291.51 15 269.80 15 Route Change 260.03 15 162.49 15 192.80 15 The p-values for all independence tests are less than 0.001
these forecasts. A similar result can be depicted for leisure trips, although
282
the behavioral changes in response to warm weather forecasts do depend
283
significantly on exposure levels. In contrast, behavioral changes in relation
284
to shopping trips are more likely to be impacted by exposure levels. In the
285
case of forecasts predicting rain, snow or fog, the behavioral changes are
286
significantly affected by weather forecasts.
287
Concerning the impact of media sources of weather forecasts on changes
288
in activity-travel behavior, Table 4 shows that the results indicate that
be-289
havioral adaptations do not depend on the media source. Irrespective of the
290
type of trips and the type of forecasted weather, the likelihood that
travel-291
ers will change their activity-travel behavior is not influenced by the media
292
source. Thus, in contrast to the effect of exposure, the media source of the
293
weather forecast does not significantly affect activity-travel behavior.
294
The final weather-forecast aspect that was investigated in this study is
295
the perceived reliability of the weather forecast, which was measured on a 10
296
point scale. A dichotomization of the perceived reliability into low perceived
297
reliability (1-5) and high perceived reliability (6-10) was performed to
in-298
vestigate the impact of perceived reliability on the probability that travelers
299
will adjust their activity-travel behaviors. From Table 4, one can see that,
300
in accordance with the results regarding the media source, the perceived
re-301
liability of the weather forecast does not affect the travelers’ frequencies of
302
making changes in their activity-travel behavior. This is true for every trip
303
type and weather condition considered in this study.
Table 4: Dependence of behavioral changes on the characteristics of the weather forecast Activity Type Weather Type Exposure1 Media Source2 Reliability3
Chi P-value Chi P-value Chi P-value
Work/School Cold 6.19 0.995 12.71 0.991 2.10 0.990 Snow 14.67 0.685 10.35 0.998 12.25 0.199 Rain 7.79 0.982 18.40 0.891 2.72 0.974 Fog 13.71 0.748 18.26 0.896 6.39 0.700 Warm 11.84 0.855 11.56 0.996 3.87 0.919 Storm 13.30 0.773 20.51 0.809 6.79 0.659 Shopping Cold 21.10 0.274 9.97 0.999 7.84 0.551 Snow 43.52 0.001 13.64 0.984 6.94 0.643 Rain 31.52 0.025 10.64 0.998 10.63 0.302 Fog 46.91 <0.001 21.02 0.785 5.09 0.827 Warm 25.10 0.122 17.34 0.922 4.14 0.902 Storm 22.82 0.198 9.47 0.999 8.51 0.484 Leisure Cold 6.58 0.993 17.55 0.917 7.81 0.554 Snow 20.65 0.298 8.75 0.999 10.51 0.311 Rain 20.85 0.287 7.17 0.999 5.21 0.816 Fog 17.44 0.493 11.77 0.995 4.17 0.900 Warm 31.04 0.028 22.25 0.724 7.56 0.579 Storm 18.06 0.452 13.53 0.985 9.24 0.415
Bold italic values indicate a significant effect of the weather forecast characteristic Degrees of freedom:
1: 18 ((3 levels of exposure - 1) × (5 behavioral changes x 2 answers (Yes/No) - 1)) 2: 27 ((4 media sources - 1) × (5 behavioral changes x 2 answers (Yes/No) - 1)) 3: 9 ((2 levels of reliability - 1) × (5 behavioral changes x 2 answers (Yes/No) - 1))
4.2. Determinants of changes in activity travel behavior and the role of weather
305
forecast characteristics
306
To determine the factors that trigger changes in activity-travel behavior
307
in the presence of adverse weather conditions, and particularly, to determine
308
the role of the characteristics (exposure, media source, perceived reliability)
309
of the weather forecasts, a GEE-MNL-model was constructed. It should be
310
noted that the prioritization of the different changes in activity-travel
behav-311
ior was required, as respondents could indicate multiple changes
simultane-312
ously (see Table 1). After prioritizing the changes in activity-travel behavior,
313
the transformed (prioritized) response variables were then analyzed using the
314
proposed GEE-MNL modeling framework.
315
The explanatory variables that were considered for the analysis are listed
316
in Table 2. Three groups of explanatory variables were taken into
consid-317
eration. The first group of explanatory variables includes the key variables
318
in this study, namely the weather-forecast-related variables: the forecasted
319
weather condition, the exposure to the weather forecasts, the media source
320
and the perceived reliability of the weather forecast. In addition to these
321
weather-forecast-related variables, the explanatory variables included
differ-322
ent descriptors of the socio-demographic profile of the respondents. Finally,
323
different transport-related variables were envisaged.
324
Separate models were estimated for each type of activity purpose. Only
325
the significant factors (assuming a level of significance of 5%) were retained
326
in the final models. The level of significance was determined by examining
327
the type III score statistics. These statistics provide insight into the overall
328
effect of a variable. For instance, in the case of a categorical variable, these
329
statistics are calculated based on the different dummy variables
simultane-330
ously. Take as an example the type of weather condition. Because there are 6
331
different weather conditions considered, the simultaneous effects of 5 dummy
332
variables are assessed using this type III score. As noted in section 3, the
333
formulation of the GEE-MNL was estimated as a multivariate binary model
334
(5 binary outcomes). Therefore, the corresponding type III score statistic
335
for this variable corresponds to 25 (5×5) degrees of freedom. It should be
336
noted that the ‘no-change’-alternative was used as the reference category in
337
this MNL model.
338
From Table 5, one can note that the type of weather condition is the most
339
predominant explanatory variable in all three models. The largest share
340
of the variance in the behavioral changes is thus attributable to the type
of weather. In addition, it should be mentioned that exposure to weather
342
forecasts, media source, and perceived reliability of the weather forecast play
343
no role. After all, only the significant factors are presented in Table 5, and
344
these variables did not have a significant impact in any of the three models.
345
Table 5: Score statistics for Type III GEE-MNL analysis
Selected Work/School Shopping Leisure
Variables Chi2 DF Sign.1 Chi2 DF Sign.1 Chi2 DF Sign.1
Weather Type 272.39 25 ? ? ? 267.98 25 ? ? ? 254.44 25 ? ? ? Age 42.06 5 ? ? ? − − − −− −− 47.55 5 ? ? ? Gender − − − −− −− 17.66 5 ?? − − − −− −− Children 12.32 5 ? − − − −− −− − − − −− −− Degree − − − −− −− − − − −− −− 27.54 15 ? Profession − − − −− −− 20.82 10 ? − − − −− −− Urbanization 40.45 15 ? ? ? − − − −− −− − − − −− −− Driving License − − − −− −− 14.13 5 ? 13.05 5 ? Season ticket 16.71 5 ?? − − − −− −− − − − −− −−
− − − indicates that this variable was not incorporated in the final model 1 Significance: n.s.: p-value ≥ 0.05, ? p-value < 0.05,
?? p-value < 0.01, ? ? ? p-value < 0.001
The interpretation of the individual parameters of weather effects, which
346
are presented in Table 6, is not straightforward. On the one hand, these
347
parameters are alternative, specific, and conditional upon the reference
al-348
ternative (i.e., the no-change alternative). On the other hand, the parameters
349
are conditional upon the reference category of the weather variable itself (i.e.,
350
extreme warm weather).
351
Concerning work- and school-related trips, it appears that fewer mode
352
changes are made in the presence of fog and cold temperatures compared
353
to snow and warm weather. The likelihood of changing the timing of the
354
trip appears to be higher for all weather conditions in comparison to warm
355
weather. Location changes appear to be least dependent on weather type;
356
nonetheless, the probability of changing the work or school location is higher
357
in the presence of snow and storms. Finally, trip cancelations and route
358
changes are most likely to occur in the presence of snow.
359
Regarding shopping trips, one could observe that mode changes are more
360
likely to be made in warm weather compared to other extreme weather
con-361
ditions. Furthermore, it is noteworthy that the timing of shopping trips is
362
most likely to be changed in the presence of rain and storms and that the
363
probability of altering the shopping location is highest during periods of fog
Table 6: Parameter Estimates for the GEE-MNL Models
Mode Time-of-day Location Cancelation Route
Parameter Est. Std.E. Est. Std.E. Est. Std.E. Est. Std.E. Est. Std.E. Parameter Estimates for the Work/School Model
Intercept -1.831 0.531 -3.260 0.531 -1.499 1.375 -1.685 0.581 -5.269 0.657 Weather Cold -1.204 0.241 1.255 0.356 0.382 0.390 -1.114 0.290 1.821 0.479 Snow 0.195 0.221 2.195 0.356 1.406 0.446 0.960 0.199 2.860 0.481 Rain -0.122 0.249 2.149 0.348 0.699 0.479 -0.725 0.240 2.145 0.481 Fog -1.238 0.324 2.233 0.353 0.314 0.583 -1.177 0.324 2.248 0.478 Storm -0.395 0.234 1.861 0.358 0.994 0.345 -0.501 0.238 1.925 0.483 Age -0.024 0.008 -0.024 0.006 -0.058 0.016 -0.016 0.008 -0.015 0.008 Children 0.686 0.274 0.322 0.261 -0.409 0.453 -0.302 0.294 -0.314 0.266 Urbanization Metropolitan 1.029 0.398 -0.762 0.367 -0.800 0.732 -0.178 0.443 0.781 0.427 Strong -0.001 0.433 -0.339 0.359 -1.830 0.705 0.919 0.491 0.793 0.438 Medium 0.365 0.356 -0.511 0.281 -0.143 0.661 0.326 0.429 0.902 0.324 Season ticket -0.129 0.229 0.381 0.234 0.004 0.352 -0.156 0.273 1.010 0.315
Parameter Estimates for the Shopping Model
Intercept -1.423 0.275 -2.660 0.385 -3.281 0.314 -2.675 0.247 -3.914 0.723 Weather Cold -1.099 0.303 0.615 0.319 0.010 0.203 -0.303 0.186 0.127 0.291 Snow -1.385 0.322 -0.006 0.302 0.595 0.248 2.202 0.159 1.062 0.847 Rain -1.180 0.289 0.725 0.282 0.633 0.232 1.471 0.158 -0.435 0.971 Fog -1.800 0.399 0.555 0.312 0.638 0.195 0.791 0.152 0.841 0.537 Storm -1.290 0.305 0.785 0.285 0.338 0.206 1.335 0.151 -2.522 1.258 Gender -0.623 0.306 -0.156 0.274 0.612 0.284 0.523 0.185 -0.801 0.590 Profession Active -0.128 0.301 -0.704 0.278 0.190 0.275 0.591 0.217 -0.833 0.440 Inactive -0.666 0.431 -0.930 0.634 0.666 0.458 0.826 0.382 0.425 0.836 Driving license -1.466 0.435 -0.117 0.421 -0.061 0.391 0.775 0.393 -0.988 0.845
Parameter Estimates for the Leisure Model
Intercept -0.031 0.412 -3.120 0.518 -1.660 0.442 -2.677 0.382 -7.199 1.041 Weather Cold -1.580 0.277 -0.542 0.587 -0.191 0.197 0.230 0.167 1.718 0.816 Snow -1.603 0.262 0.408 0.476 -0.227 0.245 2.177 0.166 2.984 0.858 Rain -1.612 0.269 1.117 0.402 0.011 0.227 1.321 0.148 2.762 0.853 Fog -2.135 0.300 0.904 0.460 -0.036 0.216 0.904 0.150 2.853 0.802 Storm -1.632 0.272 0.415 0.479 0.117 0.205 1.383 0.152 2.128 0.888 Age -0.054 0.008 -0.009 0.011 -0.020 0.011 0.025 0.008 -0.020 0.012 Degree No secondary 1.832 0.524 0.091 0.685 0.401 0.647 -0.763 0.422 2.617 0.684 Secondary 0.367 0.354 0.041 0.446 0.309 0.374 -0.211 0.256 1.172 0.570 College 0.318 0.398 -0.561 0.455 -0.363 0.405 -0.109 0.263 2.124 0.595 Driving license -0.550 0.346 -1.236 0.972 0.108 0.440 0.136 0.471 -2.155 0.787
or rain. Finally, in accordance with commuting trips, trip cancelations and
365
route changes have the highest likelihood of occurring in the presence of
366
snow.
367
The investigation of the parameter estimates relating to leisure trips
368
of switching the transport mode is the highest in warm weather. In addition,
370
one could observe that rain and fog are associated with the highest
probabil-371
ity to make time-of-day changes. Changes in the leisure location do not to
372
appear to be related to the type of weather. Finally, similar to commuting
373
and shopping trips, trip cancelations and route changes are more likely to
374
occur in the presence of snow.
375
With regard to the socio-demographic profile of the respondents, it is
376
clear from Table 5 that various socio-demographic variables contribute to
377
explaining the changes in activity-travel behavior. However, the
contribu-378
tions of these variables are limited to specific types of trips. Take age as
379
an example. With the exception of cancelations of leisure trips, for which
380
age has an increasing effect, age only has a significant effect on the
likeli-381
hood of making changes in work/school and leisure trips. In particular (see
382
Table 6), higher ages correspond to a lower likelihood of making behavioral
383
adaptations in these types of trips. In contrast, the probability of adapting
384
shopping trips is not influenced by age.
385
With respect to transport- and travel-related attributes, it may be that
386
only the possession of a driving license and a season ticket for public
trans-387
port play roles. The ownership of various transport modes and the frequency
388
of making trips with a certain activity purpose are not influential. The
pos-389
session of a season ticket appears to decrease the likelihood of route changes
390
during commuting trips. Note that the sign of the estimate – the estimate
391
corresponds to the respondents without season ticket – is positive.
There-392
fore, the effect of having a season ticket is negatively associated with the
393
probability of altering routes. Similarly, one could predict that the
posses-394
sion of a driving license increases the chance of changing transport modes in
395
shopping trips and making route changes during leisure trips. In contrast,
396
the likelihood of canceling shopping trips is lower for persons that possess a
397
driving license
5. Discussion
399
The results presented in the previous section underscore the dual role
400
of weather forecasts regarding changes in activity-travel behaviors. On the
401
one hand, the results from both the independence tests (Table 3) and the
402
GEE-MNL-model (Table 5) clearly illustrate the following significant effect
403
of forecasted weather: the likelihood of making changes in activity-travel
404
behavior depends significantly on the type of weather. On the other hand,
405
the different methods of acquiring weather information (exposure, media
406
sources, and perceived reliability) did not appear to impact the probability
407
of behavioral adaptations.
408
This aforementioned duality is partially related to the difference(s)
be-409
tween weather forecasts and the true traffic and roadway conditions. It is
410
more difficult for travelers to assess the effects of weather forecasts on road
411
weather conditions, particularly with regard to their own observations of
412
weather conditions. Often, behavioral alterations based on the travelers’ own
413
weather perceptions are limited to last-minute adaptations, such as
chang-414
ing the route or making time-of-day changes. Other adjustments, such as
415
changing the activity location, changing transport mode or canceling the
416
trip/activity, typically require longer times because these adaptations are
417
generally planned more ahead of time and, thus, fall out of this last-minute
418
range. This is the case for all three types of trips but is especially true for
419
commuting trips. This observation is also underscored by the descriptive
420
analysis (Figure 2), which showed that last minute alterations are, by far,
421
more often chosen in response to adverse weather conditions than so-called
422
’planned’ changes. To encourage ’planned’ adaptations, the discrepancies
423
between weather forecasts and the actual road conditions should be reduced.
424
One possibility is to link road weather information systems to weather
fore-425
casts. Studies from Kilpel¨ainen and Summala (2007) and Sihvola (2009)
426
showed that such road weather information services and, thus, weather
fore-427
casts have clear effects on trip schedulers. Nonetheless, the challenge lies in
428
tailoring such a system for the specific context of Flanders and the Flemish
429
weather.
430
A concern that is often raised with regard to stated adaptation
experi-431
ments is the validity of the results. In this regard, it is important to stress
432
that all the results presented in this paper were weighted such that there was
433
an optimal correspondence between the true population and the respondents
434
of the survey. Moreover, an internal validity check was performed to assess
the quality of the data. Table 7 presents the independence tests of both the
436
impact of actual weather and that of forecasted weather on the changes in
437
activity-travel behavior. As expected, the effect of actual weather conditions
438
is greater than the influence of forecasted weather conditions. This result
439
can be observed by comparing the larger Chi2-values for actual weather
con-440
ditions with the values corresponding to forecasted weather conditions. Note
441
that such comparisons of Chi2-values can be made as long as both compared
442
values correspond to the same number of degrees of freedom. The larger
im-443
pact of actual weather, therefore, provides internal evidence that the stated
444
adaptation experiment is valid.
445
Table 7: Significance of the type of weather: actual weather vs. forecasted weather1
Activity Behavioral Actual Weather Forecasted Weather
Type Change Chi2 DF Chi2 DF
Work/School Mode Change 138.71 15 108.95 15 Postpone Trip 409.05 15 353.43 15 Location Change2 81.12 15 51.51 5 Cancelation Trip2 174.79 5 111.28 5 Route Change 362.56 15 260.03 15 Shopping Mode Change 92.24 15 57.58 15 Postpone Trip 542.97 15 290.50 15 Location Change 235.69 15 101.01 15 Cancelation Trip 555.65 15 291.51 15 Route Change 302.34 15 162.49 15 Leisure Mode Change 107.92 15 80.03 15 Postpone Trip 522.45 15 311.48 15 Location Change 62.85 15 42.55 15 Cancelation Trip 405.26 15 269.80 15 Route Change 357.76 15 192.80 15 1 All p-values < 0.001
2 Estimated using reduced answer possibilities (Yes/No)
6. Conclusions
446
This paper accurately assesses how weather forecasts induce changes in
447
Flemish activity-travel behavior. The most important result is the dual role
448
of weather forecasts with regard to activity-travel behavior. This duality
pro-449
vides insight into the ambiguity of the findings reported in the international
450
literature. Moreover, the results validate the previously published findings
of Khattak and de Palma (1997), who observed no significant effect of
ac-452
quiring forecasted weather information on the likelihood of adapting mode
453
choice and departure times.
454
The deeper understanding of how weather forecasts directly and indirectly
455
affect traffic intensities provides insight for policy makers with regard to
456
mitigating the negative impacts of forecasted adverse weather conditions.
457
Therefore, the effect(s) of weather forecasts on travel behavior in
weather-458
sensitive dynamic traffic models must be taken into consideration. These
459
types of models will lead to more accurate traffic forecasts and can serve as
460
important decision support tools for both long-term and short-term policy
461
decisions. Take, for example, the case in which traffic managers attempt
462
to reduce the negative impacts of inclement weather by intervening through
463
various weather-related advisory and control measures; this practice is also
464
referred to as weather responsive traffic management. A weather-sensitive
465
traffic model could be a useful decision support tool for determining which
466
measure is most applicable to a particular situation.
467
Furthermore, as noted in the discussion section, this study recommends
468
the implementation of a road weather information system that is directly
469
linked to weather forecasts in an attempt to address the discrepancy between
470
the weather forecasts and the traffic and roadway conditions in Flanders.
471
Future research efforts should focus on the integration of the present
472
findings into travel demand modeling frameworks. Moreover, models that
473
directly link the effects of weather forecasts to the overall traffic observed
474
on the network should be developed and further enhanced. Finally, data
475
collection methods should attempt to survey both weather conditions and
476
associated travel behavior with as much detail (both in space and time) as
477
possible.
7. Acknowledgements
479
The authors would like to acknowledge Professor Wets for his guidance.
480
In addition, the authors would like to express their gratitude to Lies Kwanten
481
for proofreading the manuscript. Finally, the authors wish to thank the two
482
anonymous reviewers for their useful comments and suggestions.
483
References
484
Agresti, A., 2002. Categorical Data Analysis. Wiley, Hoboken, NJ. second
485
edition.
486
Akar, G., Clifton, K., 2009. Influence of individual perceptions and
bicy-487
cle infrastructure on decision to bike. Transportation Research Record:
488
Journal of the Transportation Research Board 2140, 165–172.
489
Ben-Elia, E., Erev, I., Shiftan, Y., 2008. The combined effect of information
490
and experience on drivers’ route-choice behavior. Transportation 35, 165–
491
177.
492
Casas, I., Kwan, M.P., 2007. The impact of real-time information on choices
493
during the commute trip: Evidence from a travel simulator. Growth and
494
Change 38, 523–543.
495
Chorus, C.G., Molin, E.J.E., Van Wee, B., 2006a. Use and effects of advanced
496
traveller information services (atis): A review of the literature. Transport
497
Reviews 26, 127–149.
498
Chorus, C.G., Molin, E.J.E., Van Wee, B., Arentze, T.A., Timmermans,
499
H.J.P., 2006b. Responses to transit information among car-drivers:
Regret-500
based models and simulations. Transportation Planning & Technology 29,
501
249–271.
502
Chung, E., Ohtani, O., Kuwahara, M., 2005. Effect of rainfall on travel time
503
and travel demand, in: Proceedings of the 5th European Congress and
504
Exhibition on Intelligent Transport Systems and Services, ERTICO - ITS
505
Europe, Hannover, Germany.
506
Cools, M., Brijs, K., Tormans, H., Moons, E., Janssens, D., Wets, G., 2011.
507
The socio-cognitive links between road pricing acceptability and changes
in travel-behavior. Transportation Research Part A: Policy and Practice
509
45, 779–788.
510
Cools, M., Moons, E., Creemers, L., Wets, G., 2010a. Changes in travel
511
behavior in response to weather conditions: Do type of weather and trip
512
purpose matter? Transportation Research Record: Journal of the
Trans-513
portation Research Board 2157, 22–28.
514
Cools, M., Moons, E., Wets, G., 2010b. Assessing the impact of public
hol-515
idays on travel time expenditure: Differentiation by trip motive.
Trans-516
portation Research Record: Journal of the Transportation Research Board
517
2157, 29–37.
518
Cools, M., Moons, E., Wets, G., 2010c. Assessing the impact of weather on
519
traffic intensity. Weather, Climate, and Society 2, 60–68.
520
Guo, Z., Wilson, N., Rahbee, A., 2007. Impact of weather on transit
rider-521
ship in chicago, illinois. Transportation Research Record: Journal of the
522
Transportation Research Board 2034, 3–10.
523
Hagens, A., 2005. De auto laten staan: ook als het regent? De invloed van
524
weer op de stedelijke verkeersvraag. Master’s thesis. University of Twente.
525
Enschede.
526
Joh, C.H., Lee, B., Bin, M., Arentze, T., Timmermans, H., 2011. Exploring
527
the use of travel information - identifying contextual market segmentation
528
in seoul, korea. Journal of Transport Geography 19, 1245–1251.
529
Khattak, A.J., de Palma, A., 1997. The impact of adverse weather
condi-530
tions on the propensity to change travel decisions: A survey of brussels
531
commuters. Transportation Research Part A: Policy and Practice 31, 181–
532
203.
533
Kilpel¨ainen, M., Summala, H., 2007. Effects of weather and weather forecasts
534
on driver behaviour. Transportation Research Part F: Traffic Psychology
535
and Behaviour 10, 288–299.
536
Kim, J., Mahmassani, H.S., Dong, J., 2010. Likelihood and duration of
537
flow breakdown: Modeling the effect of weather. Transportation Research
538
Record: Journal of the Transportation Research Board 2188, 19–28.
Koetse, M.J., Rietveld, P., 2009. The impact of climate change and weather
540
on transport: An overview of empirical findings. Transportation Research
541
Part D: Transport and Environment 14, 205–221.
542
Kuhnimhof, T., Chlond, B., Huang, P.C., 2010. Multimodal travel choices of
543
bicyclists: Multiday data analysis of bicycle use in germany.
Transporta-544
tion Research Record: Journal of the Transportation Research Board 2190,
545
19–27.
546
Kuss, O., McLerran, D., 2007. A note on the estimation of the multinomial
547
logistic model with correlated responses in sas. Computer Methods and
548
Programs in Biomedicine 87, 262–269.
549
Lam, W.H., Shao, H., Sumalee, A., 2008. Modeling impacts of adverse
550
weather conditions on a road network with uncertainties in demand and
551
supply. Transportation Research Part B: Methodological 42, 890–910.
552
Lyons, G., 2006. The role of information in decision-making with regard to
553
travel. Intelligent Transport Systems, IEE Proceedings 153, 199–212.
554
Madre, J.L., Axhausen, K., Br¨og, W., 2007. Immobility in travel diary
555
surveys. Transportation 34, 107–128.
556
Maze, T., Agarwal, M., Burchett, G., 2006. Whether weather matters to
557
traffic demand, traffic safety, and traffic operations and flow.
Transporta-558
tion Research Record: Journal of the Transportation Research Board 1948,
559
170–176.
560
Miller, M.E., Davis, C.S., Landis, J.R., 1993. The analysis of longitudinal
561
polytomous data: Generalized estimating equations and connections with
562
weighted least squares. Biometrics 49, 1033–1044.
563
de Palma, A., Rochat, D., 1999. Understanding individual travel decisions:
564
results from a commuters survey in geneva. Transportation 26, 263–281.
565
Rodr´ıguez, D.A., Levine, J., Agrawal, A.W., Song, J., 2011. Can
infor-566
mation promote transportation-friendly location decisions? a simulation
567
experiment. Journal of Transport Geography 19, 304–312.
Sihvola, N., 2009. Driver assessment of road weather conditions and road
569
weather information, in: Proceedings of the European Conference of
Trans-570
port Research Institutes (ECTRI) Young Researchers Seminar 2009,
EC-571
TRI, Torino, Italy.
572
Son, S., Khattak, A., Chen, J.Y., 2011. Comparative analysis of
univer-573
sity students’ acquisition and use of travel information. Transportation
574
Research Record: Journal of the Transportation Research Board 2243,
575
46–54.
576
Sumalee, A., Uchida, K., Lam, W.H., 2011. Stochastic multi-modal
trans-577
port network under demand uncertainties and adverse weather condition.
578
Transportation Research Part C: Emerging Technologies 19, 338–350.
579
Tseng, Y.Y., Knockaert, J., Verhoef, E.T., 2012. A revealed-preference
580
study of behavioural impacts of real-time traffic information.
Trans-581
portation Research Part C: Emerging Technologies. In press, doi:
582
10.1016/j.trc.2011.11.006.
583
Wang, X., Khattak, A., Fan, Y., 2009. Role of dynamic information in
sup-584
porting changes in travel behavior: Two-stage process of travel decision.
585
Transportation Research Record: Journal of the Transportation Research
586
Board 2138, 85–93.
587
Wilton, R.D., P´aez, A., Scott, D.M., 2011. Why do you care what other
588
people think? a qualitative investigation of social influence and
telecom-589
muting. Transportation Research Part A: Policy and Practice 45, 269–282.
590
Zhang, L., Levinson, D., 2008. Determinants of route choice and value of
591
traveler information: A field experiment. Transportation Research Record:
592
Journal of the Transportation Research Board 2086, 81–92.
